7RADIATIVE PROCESSES
Introduction. It was established in §4 of the preceding chapter that the leadingterm on the right side of (459/461)—the acceleration-independent term thatfalls off as 1/r2—admits straightforwardly of interpretation as the Coulombfield of the source, as seen from the field point, where the phrase “as seen from”alludes to
• a “retardation effect:” the field point senses not the “present location”of the source (a notion that relativity declares to be meaningless) butthe location of the puncture point—the point at which the worldline ofthe source punctured the lightcone that extends backward from the fieldpoint (a notion that does make relativistic good sense);
• the fact that if the field point sees the source to be moving at the momentof puncture then it sees not the familiar “Coulomb field of a charge atrest” but a Lorentz transform of that field.
We turn now to discussion of the structure and physical ramifications of theremaining term on the right side of (459/461)—the acceleration-dependent termthat falls off as 1/r1. This is physics for which elementary experience providesno sharp intuitive preparation, but which lies at the base of much that is mostcharacteristic of classical electrodynamics. The details are occasionally a bitintricate, and their theoretical/phenomenological/technological consequencesremarkably diverse . . .which is why I give the subject a chapter of its own.
1. Radiation fields. Dropping the Coulombic component from the field (459) ofa moving charge we obtain the radiation field
Fµν = e4πc2
[1r
(bµaν− bνaµ) + (wµaν− wνaµ) − (aw)(wµbν− wνbµ)
]0
372 Radiative processes
But (see again page 359)
wµ0 =
[Rµ
r− bµ
]0
bµ ≡ 1cu
µ
r ≡ 1cRαu
α = (Rb) = γ(1 − β‖)R
so after a short calculation we find
Fµν = e4π
[1
(Ru)2
(Rµaν−Rνaµ) − (Ra)(Ru)
(Rµuν−Rνuµ)]
0
= e4π
[1
(Ru)2(Rµaν
⊥−Rνaµ⊥)
]0
(464.1)
where aµ⊥ ≡ aµ − (Ra)
(Ru)uµ (464.2)
is (in the Lorentzian sense) ⊥ to Rµ: (Ra⊥) = 0. Note the manifest covarianceof this rather neat result.
3-vector notation—though contrary to the spirit of the principle of manifestcovariance, and though always uglier—is sometimes more useful. Looking backagain, therefore, to (461), we observe that279
RRR×((RRR− βββ ) × aaa
)= −(1 −RRR···βββ)
aaa− RRR···aaa
1 −RRR···βββ(RRR− βββ )
)︸ ︷︷ ︸
⊥ RRR
and that on this basis the radiative part of (461) can be written280
EEE = − e4πc2
[ 1R(1 −RRR···βββ )2
aaa− RRR···aaa
1 −RRR···βββ(RRR− βββ )
)]0
(465.1)
BBB =[RRR×EEE
]0
(465.2)
Equations (464) & (465) provide notationally distinct but physically equivalentdescriptions of the radiation field generated by an accelerated point charge.
It is instantaneously possible to have vvv = 000 but aaa = 000; i.e., for a pointmomentarily at rest to be accelerating. In such a circumstance (465.1) becomes
EEE = − e4πc2
[ 1R
aaa⊥
]0
aaa⊥ = aaa− (RRR···aaa)RRR = −RRR× (RRR× aaa)
= e4πc2
[RRR× (RRR× aaa)
]0
(466)
with consequences which are illustrated in Figures 123 & 124.
279 problem 75.280 We make use here of r ≡ γ(1 −RRR···βββ )R: see again page 359.
Radiation fields 373
Figure 123: Electric field at points that look back to the samepuncture point, where they see the charge to be momentarily atrest but accelerating (in the direction indicated by the green arrow).The red EEE-vectors arise from the radiative term (466). Addition ofthe Coulombic component produces the black EEE-vectors. The greyarrows are unit vectors RRR . The figure is deceptive in one respect:every EEE-vector on the left should, according to (466), have the samelength as its counterpart on the right.
The intricate details of (461)are well-adapted to computer-graphic analysis.In this connection every student of electrodynamics should study the classiclittle paper by R. Y. Tsien,281 from which I have taken Figures 125–128. Tsienassumes the source orbit to lie in every case in a plane, and it is in that planethat he displays the “electric lines of force.” From his figures one can read offthe direction of the retarded EEE -field, but information pertaining directly to themagnitude of the EEE -field (and all information pertaining to the BBB -field) hasbeen discarded. Nor does Tsien attempt to distinguish the radiative from theCoulombic component of EEE.
281 “Pictures of Dynamic Electric Fields,” AJP 40, 46 (1972). Computersand software have come a very long way in thirty years: the time is ripe forsomeone to write (say) a Mathematica program that would permit students todo interactively/experimentally what Tsien labored so hard to do with relativelyprimitive resources. Tsien, by the way, is today a well-known biophysicist, whoin 1972 was still an undergraduate at Harvard, a student of E. M. Purcell, whoseinfluential Electricity & Magnetism (Berkeley Physics Course, Volume II) wasthen recent.
374 Radiative processes
Figure 124: Shown below: the worldline of a charged particle—initially at rest—that begins abruptly to accelerate to the right, thenpromptly decelerates, returning again to rest. Shown above is theresulting EEE-field. The remote radial section is concentric about theoriginal position, the inner radial section is concentric about thealtered position. The acceleration-dependent interpolating field hasthe form shown in Figure 123. Indeed: it was from this figure—not(466)—that I took the details of Figure 123. The next figure speaksmore precisely to the same physics.
Radiation fields 375
Figure 125: Snapshots of electric field lines derived from theEEE-field generated by a charge which abruptly decelerates whilemoving in the → direction. The initial velocity was β = 0.20 inthe upper figure, β = 0.95 in the lower figure. I am indebted toFred Lifton for the digitization of Tsien’s figures, and regret thatthe available technology so seriously degraded the quality of Tsien’swonderfully sharp images. See the originals in Tsien’s paper 264. . . orbetter: run Tsien’s algorithm on Mathematica to produce animatedversions of the figures.
376 Radiative processes
Figure 126: Snapshots of the electric field lines generated by acharge undergoing simple harmonic motion in the direction. Inthe upper figure βmax = 0.10, in the middle figure βmax = 0.50, inthe lower figure βmax = 0.90.
Radiation fields 377
Figure 127: Snapshots (not to the same scale) of the electric fieldlines generated by a charge undergoing uniform circular motion about the point marked •. In the upper figure β = 0.20, in the lowerfigure β = 0.50. In the upper figure the field is—pretty evidently—dominated by the Coulombic component of (459/461).
378 Radiative processes
Figure 128: Enlargement of the same physics as that illustrated inFigure 127, except that now β = 0.95. The figure can be animatedby placing it on a phonograph turntable: since phonographs turn the spiral will appear to expand. Beyond a certain radius thefield lines will appear to move faster than the speed of light. Thatviolates no physical principle, since the field lines themselves arediagramatic fictions: marked features of the field (for example: thekinks) are seen not to move faster than light. At such high speedsthe field is dominated by the radiative part of (459/461). This is“synchrotron radiation,” and (as Tsien remarks) the kinks accountfor the rich harmonic content of relativistic synchrotron radiation.
2. Energetics of fields produced by a single source. To discuss this topic allwe have in principle to do is to introduce (459/461)—which describe the fieldgenerated by a point charge in arbitrary motion—into (309, page 215)—whichdescribes the stress/energy/momentum associated with an arbitrarily prescribedelectromagnetic field . The program is clear-cut, but the details can easilybecome overwhelming . . . and we are forced to look only at the physically mostcharacteristic/revealing features of the physically most important special cases.
Energetics 379
The experience thus gained will, however, make it relatively easy to thinkqualitatively about more realistic/complex problems.
We will need to know (see again page 216) that
E = 12 (E2 + B2)
SSS = c(EEE ×BBB )PPP = 1
c (EEE ×BBB )
describes energy density
describes energy flux
describes momentum density
but will have no direct need of the other nine components T of the stress-energytensor Sµν . Mechanical properties of the fields generated by accelerated sourceslie at the focal point of our interest, but to place that physics in context welook first to a couple of simpler special cases:
field energy/momentum of a charge at rest In the rest frameof an unaccelerated charge e we have
EEE = e4π
1R2
RRR and BBB = 000
giving
E = 12
(e4π
)2 1R4
and SSS = PPP = 000
If (as in problem 10) we center a (mental) sphere of radius a on the charge wefind the field energy exterior to the sphere to be given by
W (a) =∫ ∞
aE(R)4πR2 dR = e2
8πa(467)
. . .which—“self-energy problem”—becomes infinite as a ↓ 0, and which whenwe set
= mc2
gives rise to the “classical radius” a = e2/8πmc2 of the massive point charge e.
field energy/momentum of a charge in uniform motion Drawingnow upon (463) we have
EEE = e4πγ2
1(1 − β2 sin2 α
) 32
1R2
RRR and BBB = βββ×EEE
But B2 = (βββ×EEE )···(βββ×EEE ) = (βββ ···βββ)(EEE ···EEE) − (βββ ···EEE)2 = β2E2 sin2 α, so
E = 12
(e
4πγ2
)2 1 + β2 sin2 α(1 − β2 sin2 α
)31R4
The momentum density PPP = 1c (EEE ×BBB ) is oriented as shown in the first of the
following figures. From
P2 = 1c2
(EEE ···EEE)(BBB ···BBB) − (EEE ···BBB)2
= 1
c2E2B2 = 1
c2β2E4 sin2 α
380 Radiative processes
EEE
R
PPP
αβββ
Figure 127: The solinoidal BBB field is up out of page at the pointshown, so PPP = 1
c (EEE ×BBB ) lies again on the page. Only P‖—thecomponent parallel to βββ—survives integration over all of space..
a
r(α) α
βββ
γ –1a
Figure 130: Lorentz contracted geometry of what in the rest frameof the charge was the familiar “sphere of radius a,” exterior to whichwe compute the total energy and total momentum. The figure isrotationally symmetric about the βββ-axis.
Energetics 381
we find that the magnitude of P is given by
P = 1cβE
2 sinα = 1cβ
(e
4πγ2
)2 1(1 − β2 sin2 α
)31R4
Turning now to the evaluation of the integrated field energy and field momentumexterior to the spherical region considered previously—a region which appearsnow to be Lorentz contracted (see the second of the figures on the precedingpage)—we have
W =∫ π
0
∫ ∞
r(α)E · 2πR2 sinαdRdα (468.1)
and PPP = Pβββ with
P =∫ π
0
∫ ∞
r(α)P sinα · 2πR2 sinαdRdα (468.2)
where r(α), as defined by the figure, is given282 by
r(α) = aγ
1√1 − β2 sin2 α
The R -integrals are trivial: we are left with
W = π(
e4πγ2
)2 γ
a
∫ π
0
1(1 − β2 sin2 α)
52
sinα + β2 sin3 α
dα
P = β
c2π
(e
4πγ2
)2 γ
a
∫ π
0
1(1 − β2 sin2 α)
52
sin3 αdα
Entrusting the surviving integrals to Mathematica, we are led to results thatcan be written283
W =(1 − 1
4γ2
)· γMc2 (469.1)
PPP = γMvvv (469.2)
with M ≡ 43
e2
8πac2= 4
3m (470)
282 The argument runs as follows: we have
x2
(a/γ)2+ y2
a2= 1 whence γ2(r cosα)2 + (r sinα)2 = a2
Divide by γ2 and obtain
r2(1 − sin2 α) + (1 − β2)r2 sin2 α = (a/γ)2
Simplify, solve for r.283 problem 76.
382 Radiative processes
The curious velocity-dependent factor
(1 − 1
4γ2
)=
34 : β = 0
1 : β = 1
Were that factor absent (which is to say: in the approximation that(1− 1
4γ2
)∼1)
we would have
P 0 ≡ 1cW = 4
3m · γc and PPP = 43m · γvvv
which (see again (276) page 193) we recognize to be the relativistic relationshipbetween the energy and momentum of a free particle with mass 4
3m. This factinspired an ill-fated attempt by M. Abraham, H. Poincare, H. A. Lorentz andothers (∼ , immediately prior to the invention of relativity) to develop an“electromagnetic theory of mass,”284 distant echos of which can be detected inmodern theories of elementary particles. We note in passing that
• (469.1) gives back (467) in the limit v ↓ 0: the 34 neatly cancels the
curious 43 , which would not happen if (on some pretext) we yielded to the
temptation to drop the otherwise unattractive(1 − 1
4γ2
)-factor.
• Equations (469) and (467) are not boost-equivalent:
(W/cPPP
)= /\\\(vvv)
(mc ≡ e2/8πac
000
)
The reason is that P 0 ≡ W/c and PPP arise by integration from a subsetSµ0 of the sixteen components of the Sµν tensor, and the four elementsof the subset are not transformationally disjoint from the other twelvecomponents.
• It becomes rather natural to ask: Could a more satisfactory result beachieved if we assumed that Maxwell’s equations must be modified inthe close proximity of charges? That relativity breaks down at smalldistances?
3. Energy radiated by an accelerated charge momentarily at rest. It is in theinterest mainly of analytical simplicity that we now assume vvv = 000, a conditionthat (when aaa = 000 ) can hold only instantaneously. But the calculation is lessartificial than might at first appear: it leads to results that are nearly exact inthe non-relativistic regime v c .
284 For a good general review—with bibliography—see R. L. Dendy, “A historyof the Abraham–Lorentz electromagnetic theory of mass” (Reed College, ).See also Chapter 2 in F. Rohrlich, Classical Charged Particles () andR. P. Feynman’s Lectures on Physics (), Volume II, Chapter 28.
Larmor’s formula 383
Borrowing now from (461) we have (set βββ = 000 )
EEE = e4π
[1
R2RRR
]0
+ e4πc2
[1R
RRR ×(RRR × aaa)]0≡ EEE C + EEE R
BBB =[RRR ×EEE
]0≡ BBB C + BBB R
where the superscript C identifies the “Coulombic component,” and R the“radiative component.” We want to study energy loss (radiation from thevicinity of the charge) so we look not to E or PPP but the energy flux vector
SSS = c(EEE ×BBB )
= SSS CC + SSS CR + SSS RC + SSS RR where
SSS CC ≡ c(EEE C ×BBB C ) ∼ 1/R4
SSS CR ≡ c(EEE C ×BBB R ) ∼ 1/R3
SSS RC ≡ c(EEE R ×BBB C ) ∼ 1/R3
SSS RR ≡ c(EEE R ×BBB R ) ∼ 1/R2
SSS CC, SSS CR and SSS RC may be of importance—even dominant importance—in the“near zone,” but they fall off faster than geometrically : only SSS RR can pertainto the “transport of energy to infinity”—the process of present concern. We
look therefore to
SSS RR = c(EEE R ×BBB R ) (471)
with BBB R =[RRR×EEE R
]0
EEE R = e4πc2
[1R
RRR ×(RRR × aaa)]0
Clearly RRR···EEE = 0 so EEE× (RRR ×EEE) = (EEE ···EEE)RRR − (RRR···EEE )EEE gives285
SSS =SRRR
S = c(EEE ···EEE ) = 14πc3
(e2
4π
)(aR
)2
sin2 ϑ (472)
where ϑ ≡ (angle between RRR and aaa). The temporal rate at which field energyis seen ultimately to stream through the remote surface differential dσdσdσ is givenby dP = SSS ···dσdσdσ. But dΩ ≡ R−2RRR···dσdσdσ is just the solid angle subtended (at e) bydσdσdσ. We conclude that the power radiated into the solid angle dΩ is given by
dP =
14πc3
(e2
4π
)a2 sin2 ϑ
︸ ︷︷ ︸ dΩ (473)
|—so-called “sine squared distribution”
The “sine squared distribution” will be shown to be characteristic of dipoleradiation, and has the form illustrated in the first of the following figures.
285 problem 77. Here and henceforth I drop the superscripts R.
384 Radiative processes
ϑ
aaa
Figure 131: The “sine squared distribution” arises when vvv ∼ 000but aaa = 000. The distribution is axially symmetric about the aaa-vector,and describes the relative amounts of energy dispatched in variousϑ-directions. The radiation is predominantly ⊥ to aaa.
Integrating over the “sphere at infinity” we find the instantaneous total radiatedpower to be given by286
P = 14πc3
(e2
4π
)a2 · 2π
∫ π
0
sin2 ϑ dϑ = 23
(e2
4π
)a2
c3(474)
This is the famous Larmor formula, first derived by Joseph Larmor in . Thefollowing figure schematizes the physical assumptions which underlie (474). Wenote that while energy may also be dispatched into the solid angle dΩ by theSSS CC, SSS CR and SSS RC it is attenuated too rapidly to contribute to the net “energyflux across the sphere at infinity.”
From the c−3-dependence of PLarmor we conclude that it is not easy toradiate. Finally, I would emphasize once again that we can expect Larmor’sformula to pertain in good approximation whatever the non-relativistic (!)motion of the source.
4. Energy radiated by a charge in arbitrary motion. When one turns to thegeneral case the basic strategy (study SSS RR in the far zone) is unchanged, butthe details287 become a good deal more complicated. In the interests of brevity
286 problem 78.287 See classical radiation (), pages 558–571.
Larmor’s formula 385
R ∼ ∞
aaa
Figure 132: Above: representation of the sine-squared radiationpattern produced by a charge seen (below) at the moment of punctureto have vvv ∼ 000 but aaa = 000.
386 Radiative processes
βββ
dΩ
RRR
Figure 133: A charged particle e pursues an arbitrary path inphysical 3-space. We are concerned with the energy radiated intothe solid angle dΩ identified by the direction vector RRR. The vectorβββ refers to the particle’s velocity at the radiative moment, and—adhering to the convention introduced in Figures 127 & 128—wewrite
α ≡ angle between RRR and βββ
No attempt has been made here to represent the instantaneousacceleration vector aaa.
and clarity I must therefore be content to report and discuss here only theresults of the detailed argument. It turns out that (see the preceding figure)an accelerated charge e radiates energy into the solid angle dΩ (direction RRR )at—in τ -time—a temporal rate given by
dP = 1(1 −RRR···βββ )5
· 14πc3
(e2
4π
)∣∣RRR ×((RRR − βββ ) × aaa
)∣∣2dΩ (475)
. . .which gives back (473) when βββ = 000.
The “Dopplerean prefactor”
D(α) ≡ 1(1 −RRR···βββ )5
= 1(1 − β cos α)5
is plotted in Figure 134. Evidently
Radiation by a charge in arbitrary motion 387
5 10 15 20 25 30
α
βββ
Figure 134: Graph of the Dopplerean factorD(α), the cross-sectionof a figure of revolution about the βββ-axis. Also shown, for purposesof comparison, is the unit circle. The figure refers to the specificcase β = 0.5.
D(α)max = D (0) = 1(1 − β)5
−→ ∞ as β ↑ 1
D(α)min = D(π) = 1(1 + β)5
−→ 132 as β ↑ 1
andD(π
2 ) = 1 : all β
We conclude that the (aaa-independent) Doppler factor serves to favor the forwardhemisphere:
Fast charges tend to throw their radiation forward.
Looking back again to (475), we see that the D(α)-factor competes with(or modulates) a factor of the form
∣∣RRR ×((RRR − βββ ) × aaa
)∣∣2. A simple argumentshows that the latter factor vanishes if and only if (RRR − βββ ) ‖ aaa . This entailsthat RRR lie in the (βββ,aaa)-plane, and that within that plane it have one or theother of the values RRR1 and RRR2 described in Figure 135. RRR1 and RRR2 describe theso-called “nodal directions” which are instantaneously radiation-free. Readingfrom the figure, we see that
• in the non-relativistic limit RRR1 and RRR2 lie fore and aft of the aaa-vector,independently (in lowest order) of the magnitude/direction of βββ : this is aproperty of the “sine squared distribution” evident already in Figure 131.
• in the ultra-relativistic limit RRR1 → βββ while RRR2 gives rise to a “danglingnote,” the location of which depends conjointly upon βββ and aaa.
From preceding remarks we conclude that the distribution function thatdescribes the rate at which a charge “sprays energy on the sphere at ∞” is (inthe general case) quite complicated. Integration over the sphere can, however,
388 Radiative processes
aaa RRR1
βββ
RRR2
Figure 135: Geometrical construction of the vectors RRR1 and RRR2
that locate the nodes of the radiative distribution in the general case.
be carried out in closed form . . . and gives rise (compare (474)) to the followingdescription of the total power instantaneously radiated by an arbitrarily movingsource :
P = −23
(e2
4π
) (aa)c3
(476)
= 23
(e2
4π
)1c3
·
γ4(aaa···aaa) + γ6(aaa···βββ)2
= · γ6
(aaa···aaa) − (aaa × βββ)···(aaa × βββ)
Equation (476) is manifestly Lorentz covariant , shows explicitly the sense inwhich Larmor’s formula (474) is a “non-relativistic approximation,” and hasbeen extracted here from the relativistic bowels of electrodynamics . . .but wasfirst obtained by A. Lienard in , only one year after the publication ofLarmor’s result, and seven years prior to the invention of special relativity !
More detailed commentary concerning the physical implications of(473–476) is most usefully presented in terms of special cases & applications. . . as below:
case aaa ‖ βββ
This is the “most favorable case” in the sense that it is parallelism (aaa×βββ = 000)that (see the last of the equations just above) maximizes P . The distribution
Radiation by a charge in arbitrary motion 389
itself can in this case be described
dPdΩ
= sin2 α(1 − β cos α)5
14πc3
(e2
4π
)aaa···aaa (477)
= D(α) ·[sine squared distribution
]
αaaa ‖ βββ
The distribution is symmetric about the (aaa‖βββ)-axis (the nodes lie fore and aft),and has the cross section illustrated below:
αmax
aaa ‖ βββ
Figure 136: Radiation pattern in the case aaa‖βββ, to be read as thecross section of a figure of revolution. The figure as drawn refersto the specific case β = 0.5. The circle has radius 1
4πc3 ( e2
4π )a2, andsets the scale. The ears of the sine squared distribution (Figure 131)have been thrown forward (independently of whether aaa is parallel orantiparallel to βββ).
The ears of the sine squared distribution (Figure 131) have been thrown forward(independently of whether aaa is parallel or antiparallel to βββ ) by action of the
390 Radiative processes
Doppler factor D(α). How much they are thrown forward is measured by
αmax = cos–1
√1 + 15β2 − 1
3β
= π
2 − 52β + 325
48 β3 − · · ·
= cos–1
4√
1 − 1516γ−2 − 1
3√
1 − γ−2
= 1
2γ−1 + 133768γ−3 + · · ·
where the former equation speaks to the non-relativistic limit β ↓ 0, and thelatter to the ultra-relativistic limit γ−1 ↓ 0. In the latter limit, the smallness ofγ−1 implies that of α: double expansion of (477)—use β =
√1 − γ−2 —gives288
dPdΩ
= a2
4πc3
(e2
4π
)32γ8
(γα)2 − 5(γα)4 + · · ·
∼ a2
4πc3
(e2
4π
)32γ8 (γα)2
[1 + (γα)2]5
case aaa ⊥ βββ
This is the “least favorable case” in the sense that it is perpendicularity thatminimizes P : reading from (476) we have (use 1 + γ2β2 = γ2 )
P = 23
(e2
4π
)a2
c3·
γ6 when aaa ‖ βββγ4 when aaa⊥βββ
Working from (475) we find that the angular distribution in the special case athand can be described
dPdΩ
= 14πc3
(e2
4π
)1
(1 −RRR···βββ )3
aaa···aaa − 1
γ2
(RRR···aaa
1 −RRR···βββ
)2= 1
4πe2
4πa2
c31
(1 − β cos α)3
1 − 1γ2
sin2 α cos2 ϕ
(1 − β cos α)2
(478)
βββ
RRR
α
ϕ
aaa ⊥ βββ
288 problems 79 & 80.
Radiation by a charge in arbitrary motion 391
βββ
aaa ⊥ βββ
Figure 137: A charge traces a circular orbit (large dashed circle)with constant speed. The figure shows a cross section of the resultingradiation pattern, which is now not a figure of revolution. Theshort dotted lines on left and right indicate the radiation-free nodaldirections, which in a 3-dimensional figure would look like dimpleson the cheeks of an ellipsoid. The small blue circle sets the scale,here as in Figure 136. The figure was extracted from (478) withϕ = 0 and, as drawn, refers to the specific case β = 0.4.
where the diagram at the bottom of the preceding page indicates the meaningsof the angles α and ϕ. Shown above is a cross section of the associated radiationpattern. Notice that the nodal directions do not lie fore and aft: both are tippedforward, and stand in an angular relationship to βββ that can be extracted fromFigure 135:
tan(angle between βββ and node) = a/β
The D(α)-factor has now enhanced the leading lobe of the radiation pattern,and attenuated the trailing lobe . . . giving rise to the “synchrotron searchlight ,”in which connection one might also look back again to Figure 128.
The radiative process just described is of major astrophysical importance(arising when electrons spiral about magnetic field lines: ) and setsa limit on the energy which can be achieved by particle accelerators of toroidalgeometry (whence the linear design of SLAC: today many of the toroidal
392 Radiative processes
accelerators scattered about the world are dedicated to the production ofsynchrotron radiation—serve, in effect, as fancy “lightbulbs”). It is thereforenot surprising that the properties of synchrotron radiation have been studiedvery closely—initially by Julian Schwinger, who asks (for example) “What arethe distinguishing spectral and polarization characteristics of the radiation seenby an observer who looks into the synchrotron beam as it sweeps past?” Fora detailed account of the theory see Chapters 39–40 in J. Schwinger et al ,Classical Electrodynamics ().
Synchrotron radiation would lead also to the
radiative collapse of the bohr atom
if quantum mechanical constraints did not intervene. To study the details ofthis topic (which is of mainly historical interest) we look specifically to the Bohrmodel of hydrogen. In the ground state the electron is imagined to pursue acircular orbit of radius289
R = 2
me2= 5.292 × 10−9cm
with velocity
v = e2
= 1
137c = 2.188 × 108cm/sec
The natural time characteristic of the system is
τ = Rv
= 3
me4= 2.419 × 10−17sec
Reproduced below is the 3rd paragraph (§1) of Bohr’s original paper (“On theconstitution of atoms and molecules,” Phil. Mag. 26,1 (1913)):
“Let us now, however, take the effect of energy radiation intoaccount, calculated in the ordinary way from the acceleration of theelectron. In this case the electron will no longer describe stationaryorbits. W will continuously increase, and the electron will approachthe nucleus describing orbits of smaller and smaller dimensions,and with greater and greater frequency; the electron on the averagegaining in kinetic energy at the same time as the whole system losesenergy. This process will go on until the dimensions of the orbit areof the same order of magnitude as the dimensions of the electronor those of the nucleus. A simple calculation shows that the energyradiated out during the process considered will be enormously greatcompared with that radiated out by ordinary molecular processes.
To make his model work Bohr simply/audaciously assumed the (classical)physical ideas thus described to be “microscopically inoperative.” But I want
289 See, for example, quantum mechanics (), Chapter 2, pages 138–139.For the duration of the present discussion I adopt rationalized units: e2/4π → e2.
Radiation by a charge in arbitrary motion 393
r
R
Figure 138: Bohr atom, in which the nuclear proton and orbitalelectron have been assigned their classical radii. We study the“collapse” of the system which would follow from classical radiationtheory if quantum mechanics did not intervene.
here to pursue the issue—to inquire into the details of the “simple calculation”to which Bohr is content merely to allude. We ask: How much energy wouldbe released by the radiative collapse of a Bohr atom, and how long would theprocess take?
If the electron and proton were literally point particles then, clearly, theenergy released would be infinite . . .which is unphysical. So (following Bohr’sown lead) let us assume the electron and proton to have “classical radii” givenby
r = e2/2mc2 and rp = r/1836.12 r
respectively, and the collapse “proceeds to contact.” The elementary physics ofKeplerean systems290 leads then to the conclusion that the energy released canbe described
E = 12e2
1
r + rp− 1R
∼ 1
2e2
1r− 1R
= 1
2e2
2mc2
e2− me
2
2
= mc2
1 − 12
(e2
c
)2∼ mc2
290 See, for example, H. Goldstein, Classical Mechanics (2nd edition ),page 97.
394 Radiative processes
The atom radiates at a rate given initially (Larmor’s formula) by
P = 23e2
c3a2
with a = v2
R=
(1
137e
)2
mc2
and has therefore a lifetime given in first approximation by
T = EP
= mc2/
23e2
c3
(1
137e
)4
(mc2)2
= 32 (137)5 τ
= (7.239 × 1010) τ
= 1.751 × 10−6sec
Despite the enormous accelerations experienced by the electron, the radiationrate is seen thus to be “small”: the orbit shrinks in a gentle spiral and the atomlives for a remarkably long time (1010 revlolutions corresponds, in terms of theearth-sun system, to roughly the age of the universe!). . .but not long enough.The preceding discussion is, of course, declared to be “naively irrelevant” bythe quantum theory (which, in the first instance, means: by Bohr) . . .whichis seen now to be “super-stabilizing” in some of its corollary effects. It can,in fact, be stated quite generally that the stability of matter is an intrinsicallyquantum mechanical phenomenon, though the “proof” of this “meta-theorem”is both intricate and surprisingly recent.291
5. Collision-induced radiation. In many physical contexts charges move freelyexcept when experiencing abrupt scattering processes, as illustrated in the figureon the facing page. We expect the energy radiated per scatter to be given inleading approximation by
Eper scatter = 23e2
4πc3(
∆vτ
)2
τ
where ∆v ≡ vout − vin and where τ denotes the characteristic duration of eachscattering event. Suppose we had a confined population of N such charges, andthat each charge experiences (on average) n collisions per unit time. We expectto have τ ∼ 1/v and n ∼ v. The rough implication is that the populationshould radiate at the rate
P ∼ NnEper scatter ∼ (∆v)2v2
If we could show that ∆v (∼ momentum transfer per collision) is v-independentwe would (by v2 ∼ temperature) have established the upshot of Newton’s law ofcooling . The point I want to make is that radiative cooling is a (complicated)radiative process. The correct theory is certainly quantum mechanical (andprobably system-dependent), but the gross features of the process appear to bewithin reach of classical analysis. A much more careful account of the radiationproduced by impulsive scattering processes can be found in Chapter 37 of theSchwinger text cited on page 392.
291 See F. J. Dyson & A. Lenard, “Stability of matter. I,” J. Math. Phys. 8,423 (1967) and subsequent papers.
The self-interaction problem 395
Figure 139: Worldline of a charged particle subject to recurrentscattering events. Brackets mark the intervals during which theparticle is experiencing non-zero acceleration.
We have concentrated thus far mainly on single-source radiative processes,though the theory of cooling invited us to contemplate the radiation producedby random populations of accelerated charges. And we will want later to studythe radiation produced when multiple sources act in concert (as in an antenna).But there are some important aspects and manifestations of single-sourceradiation theory which remain to be discussed, and it is to these that I nowturn.
6. The self-interaction problem. We know that charges feel—and accelerate inresponse to—impressed electromagnetic fields. But do charges feel their ownfields?. . . as (say) a motorboat may interact with the waves generated by its ownformer motion? Thought about the dynamics of a free charge at rest makes itappear semi-plausible that charges do not feel their own Coulomb fields. But thesituation as it pertains to radiation fields is much less clear . . . for when a charge“radiates” it (by definition) “mails energy/momentum to infinity” and thusacquires a debt which (by fundamental conservation theorems) must somehowbe paid. One might suppose that the responsibility for payment would fall to theagency which stimulated the charge to accelerate. But theoretical/observationalarguments will be advanced which suggest that there is a sense in whichaccelerated charges do feel—and recoil from—their own radiative acts.
396 Radiative processes
Compton length /mc
Figure 140: The “classical electron” • is not, as one might expect,larger than but much smaller than the “quantum electron.” A photonwith wavelength λ = e2/2mc2 short enough to permit one to see the• would carry energy E = hν = hc/λ = (hc/e2)2mc2 = 137 · 2mc2enough to create 137 electron-positron pairs . . . and in the clutterthe intended object of the measurement process would be lost!
The point at issue is made complicated by at least three interrelatedcircumstances. The first stems from the fact that the structural propertieswhich distinguish “radiation fields” become manifest only in the “far zone,”but it is in the “near zone” that (in a local theory like electrodynamics) anyparticle/self-field interaction must occur . The second derives from the truismthat “to describe the motorboat-wake interaction one must know somethingabout the geometry of motorboats”: similarly, to study the electrodynamicalself-interaction problem one must be prepared to make assumptions concerningthe “structure oif charged particles.” Classical theory speaks of “point particles”and—in the next breath—of “charged balls” of classical radius e2/2mc2, but (asAbraham/Lorentz/Poincare discovered: see again page 382) seems incapable ofgenerating a seriously-intended electron model. Which is hardly surprising, forelectrons (and charged particles generally) are quantum mechanical objects. In
The self-interaction problem 397
this connection it is illuminating to note that the “quantum radius” of a masspoint is (irrespective of its charge) given by /mc. But
“classical radius” ≡ e2
mc2= e2
c·
mc= “quantum radius”
137
. . . so the “classical electron” is much smaller than the “quantum electron.”292
Which brings us to the third complicating circumstance (Figure 140): we seeka classical theory of processes which are buried so deeply within the quantumregime as to make the prospects of a formally complete and self-consistent theoryseem extremely remote. From this point of view the theory described below—imperfect though it is—acquires a semi-miraculous quality.
Limited success in this area was first achieved () by M. Abraham, whoargued non-relativistically—from energy conservation. We have
FFF + FFFR = maaa where
FFF ≡ impressed force
FFFR ≡ self-force, the nature of which weseek to determine
FFF may act to change the energy of the (charged) particle, but we semi-expectFFFR to conform to the energy balance condition
(work on particle by FFFR) + (energy radiated) = 0
Drawing upon Larmor’s formula (474) we are led thus to write (on a typicaltime interval t1 t t2)
∫ t2
t1
FFFR···vvv dt+ 23
(e2
4π
)1c3
∫ t2
t1
aaa···aaa dt︸ ︷︷ ︸
= 0
Integration by parts gives
= aaa···vvv∣∣∣t2t1−
∫ t2
t1
aaa···vvv dt
If it may be assumed (in consequence of periodicity or some equivalentcondition) that
aaa···vvv∣∣∣t2t1
= 0
then ∫ t2
t1
FFFR − 2
3
(e2
4π
)1c3aaa···vvv dt = 0
292 Nor is this fact special to electrons. Since m enters identically on left andright, it pertains also to protons, to every particle species.
398 Radiative processes
This suggests—but does not strictly entail—that FFFR may have the form
FFFR = 23
(e2
4π
)1c3xxx... (479.1)
More compactly, = mτ xxx... (479.2)
where the parameter τ can be described
τ ≡ 23
(e2
4π
)1mc3
= 43
(e2
8πmc2)
1c
= 43
classical particle radiusc
∼
time required for light to transit fromone side of the particle to the other
The non-relativistic motion of a charged particle can—on the basis of theassumptions that led to (479)—be described
FFF +mτxxx... = mxxx (480.1)
or again FFF = m(xxx− τxxx...) (480.2)
. . .which is the so-called “Abraham-Lorentz equation.” This result has severalremarkable features:
• It contains—which is uncommon in dynamical contexts—an allusion tothe 3rd derivative. This, by the way, seems on its face to entail thatmore than the usual amount of initial data is required to specify a uniquesolution.
• The Abraham-Lorentz equation contains no overt allusion to particlestructure beyond that latent in the definition of the parameter τ .
• The “derivation” is susceptible to criticism at so many points293 as tohave the status of hardly more than a heuristic plausibility argument.It is, in this light, interesting to note that the work of 75 years (bySommerfeld, Dirac, Rohrlich and many others) has done much to “cleanup the derivation,” to expose the “physical roots” of (480) . . .but hasat the same time shown the Abraham-Lorentz equation to be essentiallycorrect as it stands . . . except that
• The Abraham-Lorentz equation (480) is non-relativistic, but this is aformal blemish which (see below) admits easily of rectification.
293 Most critically, the argument draws upoon the Larmor formula—a “farfield result”—to obtain information about “near field physics.” The first of the“complicating circumstances” mentioned on page 396 is not only not illumi-inated/resolved, it is not even addressed.
The self-interaction problem 399
We recall from page 192 that the 4-acceleration of a moving point can bedescribed
a(τ) ≡ d2
dτ2x(τ) =(
1cγ
4(aaa···vvv)γ2aaa+ 1
c2 γ4(aaa···vvv)vvv
)
where vvv and aaa are “garden variety” kinematic 3-variables: vvv ≡ dxxx/dt andaaa ≡ dvvv/dt. We know also (page 192/193) that
(u, a) = c2 (481.1)(u, a) = 0 (481.2)
and can sho by direct computation that
(a, a) = −γ4
(aaa···aaa) + 1c2 γ
2(aaa···vvv)2
(481.3)
while a somewhat more tedious computation gives
b(τ) ≡ ddτ a(τ) (482)
= γ3
(1c γ
2[(aaa···vvv) + (aaa···aaa) + 4 1
c2 γ2(aaa···vvv)2
]aaa+ 3 1
c2 γ2(aaa···vvv)aaa+ 1
c2 γ2[(aaa···vvv) + (aaa···aaa) + 4 1
c2 γ2(aaa···vvv)2
]vvv
)
where aaa ≡ ddtaaa = xxx...
A final preparatory computation gives
(u, b) = γ4
(aaa···aaa) + 1c2 γ
2(aaa···vvv)2
= −(a, a) (481.4)
We are in position also to evaluate (a, b) and (b, b), but have no immediate needof such information . . . so won’t.294 Our immediate objective is to proceed fromFFFR = 2
3 (e2/4π) 1c3xxx... to its “most natural” relativistic counterpart—call it Kµ
R . Itis tempting to set KR = 2
3 (e2/4π) 1c3 b, but such a result would—by (481.4)—be
inconsistent with the general requirement (see again page ???) that (K,u) = 0.We are led thus—tentatively—to set
KR = 23
(e2
4π
)1c3b⊥ (483)
b⊥ ≡ b− (b, u)(u, u)
u
= b+(a, a)c2
u
= γ3
(1c γ
2[(aaa···vvv) + 3 1
c2 γ2(aaa···vvv)2
]aaa+ 3 1
c2 γ2(aaa···vvv)aaa+ 1
c2 γ2[(aaa···vvv) + 3 1
c2 γ2(aaa···vvv)2
]vvv
)
in which connection we note that
↓=
(0FFFR
)in the non-relativistic limit (as required)
294 problem 81.
400 Radiative processes
Now, the spatial part of Minkowski’s equation Kµ = md2x/dτ2 can (see again(288) page 197) be written (1/γ)KKK = d
dt (γmvvv), and in this sense it is (not KKKbut) (1/γ)KKK which one wants to call the “relativistic force.” We are led thusfrom (483) to the conclusion that the relativistic self-force
FFFR = 23
(e2
4π
)1c3γ2
aaa+ 3 1
c2 γ2(aaa···vvv)aaa+ 1
c2 γ2[(aaa···vvv) + 3 1
c2 γ2(aaa···vvv)2
]vvv
(484.1)
This result was first obtained () by Abraham, who however argued not fromrelativity but from a marginally more physical refinement of the “derivation” of(479). The “argument from relativity” was first accomplished by M. von Laue(). The pretty notation
FFFR = 23
(e2
4π
)1c3γ4
ggg + 1
c2 vvv × (vvv × ggg)
(484.2)
ggg ≡ aaa+ 3 1c2 γ
2(aaa···vvv)aaa
was introduced into the modern literature by David Griffiths,295 but wasreportedly original to Abraham.296
All modern self-interaction theories297 hold (483)—which can be notated
KµR = 2
3
(e2
4π
)1c3
d3xµ
dτ3+ 1c2
(aαaα)dxµ
dτ
aα ≡ d2xα
dτ2
—to be exact (so far as classical theory allows). Which is surprising, for we havedone no new physics, addressed none of the conceptual difficulties characteristicof this topic. We note with surprise also that we can, in the relativistic regime,have FFFR = 000 even when aaa = 000.
To study the physical implications of the results now in hand we retreat(in the interest of simplicity) to the non-relativistic case: (480). If (also forsimplicity) we assume FFF to be xxx-independent (i.e., to be some arbitrarilyprescribed function of t alone) then the Abraham-Lorentz equation (480) reads
xxx...− 1
τ xxx = − 1mτ FFF (t) (485)
and entails
xxx(t) = et/τaaa− 1
mτ
∫ t
0
e−s/τFFF (s) ds
(486.1)↑—constant of integration
295 “Dumbbell model for the classical radiation reaction,” AJP 46 244 (1978).296 problem 82.297 For references see the Griffiths paper just cited.
The self-interaction problem 401
Successive integrations give
xxx(t) = vvv +∫ t
0
xxx(s) ds (486.2)
and
xxx(t) = xxx+∫ t
0
xxx(s) ds (486.3)
where vvv and xxx are additional constants of integration.298
In the force-free case FFF (t) ≡ 000 equations (486) promptly give
xxx(t) = xxx+ vvv t+ aaaτ2et/τ
This entails xxx(t) = vvv + aaaτet/τ , which is asymptotically infinite unless aaa = 000.So we encounter right off the bat an instance of the famous run-away solutionproblem, which bedevils all theories of self-interaction. It is dealt with byconjoining to (485) the stipulation that
Run-away solutions are to be considered“unphysical” . . . and discarded. (487)
One (not immediately obvious) effect of the asymptotic side-condition (487) isto reduce to its familiar magnitude the amount of initial data needed to specifya particular particle trajectory.
To gain some sense of the practical effect of (487) we look next to thecase of an impulsive force FFF (t) ≡ mτAAAδ(t− t0) . Immediately
xxx(t) =
et/τaaa : t < t0
et/τ[aaa−AAAe−t0/τ
]: t > t0
Thbe requirement—(487)—that xxx(t) remain asymptotically finite entails thatthe adjustable constant aaa be set equal to AAAe−t0/τ . Then
xxx(t) =
AAAe(t−t0)/τ : t < t0
000 : t > t0
(488)
The situation is illustated in Figure 141. The most striking fact to emergeis that the particle starts to accelerate before it has been kicked! This is aninstance of the famous preacceleration phenomenon. It is not an artifact of theδ-function, not a consequence of the fact that we are working at the momentin the non-relativistic approximation . . .but a systemic feature of the classicalself-interaction problem. Roughly, preacceleration may be considered to arise
298 problem 83.
402 Radiative processes
tt0
Figure 141: Graphs of (reading from top to bottom) the impulsiveforce FFF (t) ≡ mτAAAδ(t − t0) and of the resulting acceleration xxx(t),velocity xxx(t) and position xxx(t). The shaded rectangle identifies the“preacceleration interval.”
because “the leading edge of the extended classical source makes advancecontact with the force field.” The characteristic preacceleration time is—consistently with this picture—small, being given by τ (∼ 10−24 seconds foran electron). On its face, preacceleration represents a microscoptic violation ofcausality . . . and so it is, but the phenomenon lies so deep within the quantumregime as to be (or so I believe) classical unobservable in every instance.Preacceleration is generally considered to be (not a physical but) a merely“mathematical phenomenon,” a symptom of an attempt to extend classicalphysics beyond its natural domain of applicability.
We may “agree not to be bothered” by the preacceleration “phenomenon.”But preacceleration comes about as a forced consequence of implementation ofthe asymptotic condition (487) . . . and the fact that the equation of motion (485)cannot stand on its own feet, but must be propped up by such a side condition,is bothersome. Can one modify the equation of motion so as to make themake the asymptotic condition automatic?. . . so that “run-away solutions”
The self-interaction problem 403
simply do not arise? The question provokes the following formal manipulation.Let (485) be written
(1 − τD)mxxx(t) = FFF (t)
or againmxxx(t) = 1
1 − τDFFF (t) (489)
where D ≡ ddt . Recalling 1
λ =∫ ∞
0
e−λθ dθ, we presume to write
11 − τD =
∫ ∞
0
e−(1−τD)θ dθ
even though D is here not a number but a differential operator (this is heuristicmathematics in the noble tradition of Heaviside). Then
mxxx(t) =∫ ∞
0
e−θeθτDFFF (t) dθ
But eθτDFFF (t) = FFF (t+ θτ) by Taylor’s theorem, so
=∫ ∞
0
FFF (t+ θτ) dθ (490)
Notice that, since c ↑ ∞ entails τ ↓ 0, we can use∫ ∞
0
e−θ dθ = 1 to recover
Newton’s mxxx(t) = FFF (t) in the non-relativistic limit. Equation (490) states thatxxx(t) is determined by a weighted average of future force values, and thereforeprovides a relatively sharp and general characterization of the preaccelerationphenomenon—encountered thus far only in connection with a single example.Returning to that example . . . insert FFF (t) ≡ mτAAAδ(t− t0) into (490) and obtain
xxx(t) =∫ ∞
0
AAAδ(t− t0 + θτ)τdθ =
AAAe(t−t0)/τ : t < t0
000 : t > t0
We have recovered (488), but by an argument that is free from any explicitreference to the asymptotic condition. In (490) we have a formulation of theAbraham-Lorentz equation (480) in which the “exotic” features have beentranslocated into the force term . . .but we have actually come out ahead: wehave managed to describe the dynamics of a self-interacting charge by meansof an integrodifferential equation of motion that stands alone, without need ofa side condition such as (487). The general solution of (490) has, by the way,the familiar number of adjustable constants of integration, so standard initialdata serves to identify particular solutions.
If in place of the “integral representation of 1/(1 − τD)” we use
11 − τD = 1 + τD + (τD)2 + · · ·
404 Radiative processes
then in place of (490) we obtain
mxxx(t) = FFF (t) + τFFF′(t) + τ2FFF
′′(t) + · · · (491)
= Newtonian force + Radiative corrections
Equations (490) and (491) are equivalent. The latter masks preacceleration(acausality), but makes explicit the Newtonian limit.299
Having thus exposed the central issues, I must refer my readers to theliterature for discussion of the technical details of modern self-interactiontheory: this is good, deep-reaching physics, which has engaged the attention ofsome first-rate physicists and very much merits close study.300 I turn now todiscussion of some of the observable physical consequences of self-interaction:
7. Thomson scattering. An electron in a microwave cavity or laser beamexperiences a Lorentz force of the form
FFF (t) = e(EEE + 1c vvv×BBB) cosωt
↓= eEEE cosωt in the non-relativistic limit
For such a harmonic driving force (486.1) becomes
xxx(t) = eΩtaaa− e
mEEEΩ∫ t
0
e−Ωs cosωs ds︸ ︷︷ ︸
where Ω ≡ 1τ = 3
24πe2mc
3. But
= e−Ωs
Ω2 + ω2
[− Ω cosωs+ ω sinωs
]t
0
so= emEEE
Ω2 cosωt− Ωω sinωtΩ2 + ω2
+ eΩtaaa− e
mEEEΩ2
Ω2 + ω2
The asymptotic condition (487) requires that we setetc.
= 000, so after some
299 For a much more elaborate discussion of the ideas sketched above seeclassical radiation (), pages 600–605.300 F. Rohrlich’s Classical Charged Particles (), Chapters 2 & 6 andJ. D. Jackson’s Classical Electrodynamics (3rd edition ), Chapter 16 aregood places to start. See also T. Erber, “The classical theories of radiationreaction,” Fortschritte der Physik 9, 343 (1961) and G. N. Plass, “Classicalelectrodynamic equations of motion with radiative reaction,” Rev. Mod. Phys.33, 37 (1961) . . .which are excellent general reviews and provide goodbibliographies. Students should also not neglect to examine the classics: Dirac(), Wheeler-Feynman ().
Scattering by a free charge 405
incident plane wavescat
tered
radiat
ion
EEE
SSS
BBB
Figure 142: A monochromatic plane wave is incident upon a freeelectron •, which is stimulated to oscillate and therefore to radiatein the characteristic sine-squared pattern. The electron drinks energyfrom the incident beam and dispatches energy in a variety of otherdirections: in short, it scatters radiant energy. Scattering by thisclassical mechanism—by free charges—is called Thomson scattering.
elementary algebra we obtain
xxx(t) = 1√1 + (ω/Ω)2
emEEE cos(ωt+ δ) (492)
where the phase shift
δ = arctan(ω/Ω)
is the disguise now worn by the preacceleration phenomenon. We note in passingthat
↓= emEEE cosωt in the non-relativistic limit: Ω ω
It is upon (492) that the classical theory of the scattering of electromagneticradiation by free electrons—“Thomson scattering”—rests. We inquire now intothe most important details of this important process.
Using (492) in conjunction with the Larmor formula (474) we conclude thatthe energy radiated per period by the harmonically stimulated electron (see thepreceding figure) can be described
∫ T
0
P dt = 23
(e2
4π
)1c3
(emE
)2 11 + (ω/Ω)2
∫ T
0
cos2 ωt dt with T ≡ 2π/ω
=(cE2πω
)· 8π
3
(e2
4πmc2)2 1
1 + (ω/Ω)2
406 Radiative processes
On the other hand, we know from work on page 305 that the (time-averagedenergy flux or) intensity of the incident plane wave can be described I = 1
2cE2
so the energy incident (per period) upon an area A becomes
ITA = 12cE
2(2π/ω)A =(cE2πω
)·A
We conclude that
A free electron absorbs (only to re-radiate) energy from anincident monochromatic wave as though it had a cross-sectionalarea given by
σThomson = 8π3 (classical electron radius)2 · 1
1 + (ω/Ω)2
The final factor can and should be dropped: it differs from unity only if
ω Ω = 32
(4πce2
)mc2 = 205mc2
and this carries us so far into the relativistic regime that we must expect ourclassical results long since to have become meaningless. Neglect of the factoramounts to neglect of the self-interaction: it entails δ = arctan(ω/Ω) → π
2 andcauses the Thomson scattering cross-section
σThomson = 8π3
[e2/4πmc2
]2 (493)
to become ω-independent. Thomson scattering—which in the respect just notedis quite atypical—may be considered to comprise the classical limit of Comptonscattering, the relativistic quantum process diagramed below. The radiation
ωout
ωin
Figure 143: In view of the fact that Compton scattering yieldsscattered photons that have been frequency-shifted it is remarkablethat no frequency shift is associated with the Thomson scatteringprocess.
Scattering by a harmonically bound charge 407
ϑ
Figure 144: Representation of the axially-symmetric sine-squaredcharacter of the Thomson scattering pattern. I invite the reader toconsider what would be the pattern if the incidentg radiation wereelliptically polarized.
field generated by a harmonically stimulated free electron has the structureillustrated in Figure 126. The differential Thomson cross-section (Figure 144)is readily seen to have the sine-squared structure
dσdΩ
∣∣∣Thomson
=[e2/4πmc2
]2 sin2 ϑ
8. Rayleigh scattering. Let our electron—formerly free—be considered now tobe attached to a spring, part of a “classical molecule.” If the spring force iswritten fff = −mω2
0xxx then the Abraham-Lorentz equation (480) becomes
xxx− τxxx...+ ω20xxx = e
mEEE cosωt (494)
We expect the solution of (494) to have (after transcients have died out) theform
xxx(t) = XXX cos(ωt+ δ)
with XXX ‖ EEE, and will proceed on the basis of that assumption—an assumptionwhich, by the way,
• renders the asymptotic condition (487) superfluous• entails xxx... = −ω2xxx.
408 Radiative processes
Our initial task, therefore, is to describe the solution
x(t) = Xei(ωt−δ)
ofx+ 2bx+ ω2
0x = emEe
iωt
b ≡ 12τω
2
But this is precisely the harmonically driven damped oscillator problem—painfully familiar to every sophomore—the only novel feature being that the“radiative damping coefficient” b is now ω -dependent. Immediately
(−ω2 + 2ibω + ω20 )︸ ︷︷ ︸ Xe−iδ = e
mE
=√
(ω20 − ω2)2 + 4b2ω2 exp
i tan–1 2bω
ω20 − ω2
which gives
X(ω) =(e/m)E√
(ω20 − ω2)2 + 4b2ω2
= eEmω2
0
1√(1 − ξ2)2 + k2ξ6
≡ eEmω2
0
X(ξ, k)
δ(ω) = tan–1 2bωω2
0 − ω2
= tan–1 kξ3
1 − ξ2 ≡ δ(ξ, k)
whereξ ≡ ω/ω0 and k ≡ τω0
are dimensionless parameters. It is useful to note that k is, in point of physicalfact, typically quite small:
k = period of optical reverberations within the classical electronperiod of molecular vibrations
∼ e2/mc3
3/me4=
(e2
c
)3
=(
1137
)3
= 3.89 × 10−7
Precisely the argument that led to (493) now leads to the conclusion that theRayleigh scattering cross-section can be described301
σRayleigh(ω) = σ0 · ω4
(ω20 − ω2)2 + 4b2ω2
(495)
= σ0ξ4
(1 − ξ2)2 + k2ξ6
σ0 ≡ σThomson = 8π3
[e2/4πmc2
]2301 problem 84.
Scattering by a harmonically bound charge 409
1
1
Figure 145: Graphs of X(ξ, k) in which, for clarity, k has beenassigned the artificially large values k = 0.15 and k = 0.05. An easycalculation shows that the resonant peak stands just to the left ofunity:∂∂ξ X(ξ, k) = 0 at ξ =
[√1 + 6k2 − 1
3k2
] 12
= 1 − 34k
2 + 6332k
4 − · · ·
and thatXmax = k–1 + 9
8k − 189128k
3 + · · ·
1 3 5 7
π
Figure 146: Graphs of δ(ξ, k) in which k has been assigned thesame artificially large values as described above. As k becomessmaller the phase jump becomes steeper, δ approaches π moreclosely, and hangs there longer before—at absurdly/unphysicallyhigh frequencies ω Ω—dropping to π
2 :
limξ↑∞
tan–1 kξ3
1 − ξ2 = limξ↑∞
tan–1(−kξ) = π2
410 Radiative processes
1 2 3 41
10
20
Figure 147: Graphs of the Rayleigh distribution function. In (495)I have set σ0 = 1 and have assigned to k the artificially large valuesk = 0.25 and k = 0.10. The red line at unity has been inserted toemphasize the high-frequency asymptote. The resonant peak lies inthe very near neighborhood of ξ ≡ ω/ω0 = 1 and its height becomesinfinite when self-interactive effects are turned off : k ↓ 0. Thephysical short of it : The apparent size of a “classical molecule”depends upon the color of the light in which it is viewed.
What we have learned is that Rayleigh scattering—energy absorption andreemission by a monochromatically stimulated and self-interactively damped“classical molecule” (charged particle on a spring)—is frequency-dependent.Looking to the qualitative details of that ω -dependence (Figure 147), we findit natural to distinguish three regimes:
low-frequency regime ξ ≡ ω/ω0 1 so with Mathematica’s aid weexpand about ξ = 0, obtaining
ξ4
(1 − ξ2)2 + k2ξ6= ξ4 + 2ξ6 + 3ξ8 + (4 − k2)ξ10 + (5 − 4k2)ξ12 + · · ·
Thus are we led to the so-called “4th power law”
σRayleigh(ω) ∼ σ0(ω/ω0)4 : ω ω0 (496)
The accuracy of the approximation is evident in Figure 148.
It is a familiar fact that (if we may allow ourselves to speak classicallyin such a connection) slight conformational/dynamical adjustments of atomic/molecular state can result in the emission (or from the absorption) of visiblelight: [∆E ≈ ∆ω0] = ω. From this we infer that the characteristic atomic/molecular vibrational frequencies ω0 are themselves than the frequencies
Scattering by a harmonically bound charge 411
0.1 0.2 0.3
0.001
0.002
0.003
0.004
Figure 148: Graph—based upon (495)—of σRayleigh with ξ 1,compared with the scattering cross-section asserted by the 4th powerlaw (496). In both cases I have set σ0 = 1, and in the former caseI have taken k = 0.00001. Naive arguments developed in the textsuggest that atomic/molecular rotational/vibrational frequencies ω0
are typically than the frequencies present in the visible spectrum.
characteristic of visible light,302 and that the scattering of sunlight by air istherefore a “low-frequency phenomenon.”303
resonance regime Here ξ ∼ 1 (i.e., ω ∼ ω0) ⇒ σ ∼ σmax andprovides a classical intepretation of the phenomenon of resonance florencence.Let (495) be written
σRayleigh = σ0ξ4
(1 + ξ)2(1 − ξ)2 + k2ξ6
≈ 14σ0
1(ξ − 1)2 + ( 1
2k)2(497)
For a comparison of the exact Rayleigh distribution function with its resonantapproximation (497), see Figure 149. The nearly Gaussian appearance of theappoximating function leads us to observe that
∫ +∞
−∞14
1(ξ − ξ0)2 + ( 1
2k)2dξ = π
2 k–1 : all ξ0
and on the basis of that information to introduce the definition
L(ξ − ξ0, k) ≡ 12π
k(ξ − ξ0)2 + ( 1
2k)2: k > 0 (498)
302 For the former we might borrow ω0 = 2π(me4/3) = 2.60 × 1017Hz fromthe Bohr theory of hydrogen (see again page 392). For visible light one has4.0 × 1014Hz < ω < 7.5 × 1014Hz.303 problem 85.
412 Radiative processes
1 2 3 41
10
20
Figure 149: Comparison of the exact Rayleigh cross-section withits resonant approximation (497). In constructing the figure I haveassigned k the unphysically large value k = 0.25. The fit—alreadyquite good—becomes ever better as k gets smaller.
We will soon (in §9) have unexpected occasion to inquire more closely intoproperties of the “Lorenz distribution function” L(ξ, k),304 but for the momentare content to observe that in this notation
σ ≈ (π/2k)σ0 · L(ξ − 1, k) at resonance: ω ∼ ω0
and that L(ξ, k) assumes its maximal value at ξ = 0: L(0, k) = 2πk
–1 so
σmax = σ0/k2 = (σ0/τ
2)/ω20 = (σ0/τ
2)/(2πν0)2 (499.1)
where ν0 is the literal frequency of the resonant radiation and (below) λ0 = c/ν0
its wavelength. But (look back again to pages 398 and 406 for the definitionsof τ and σ0)
σ0/τ2 = 8π
3
[e2/4πmc2
]2/[23e
2/4πmc3]2 = 6πc2
so
σmax = 6π(c/2πν0)2 = 3
2πλ20 (499.2)
∼
cross-sectional area of the smallest objectvisible in radiation of resonant frequency
Radiation of resonant frequency, when incident upon a “gas” made of such“classical molecules,” is scattered profusely (the gas becomes“florescent,” and
304 Also—and with better reason—called the “Cauchy distribution function.”See Abramowitz & Stegun, Handbook of Mathematical Functions (), page930.
Classical theory of spectral line shape 413
loses its transparency). Classically, we expect a molecule to possess a varietyof normal modes . . . a variety of “characteristic frequencies,” and resonanceflorescence to occur at each. Notice that if we were to neglect the self-interaction(formally: let τ ↓ 0 in (499.1)) then the resonant scattering cross-sectionwould become infinite: σmax ↑ ∞. Here as in (for example) the elementarytheory of forced damped oscillators, it is damping that accounts for finitenessat resonance.
high-frequency regime If ξ 1 then (495) becomes
σRayleigh = σ0 · 11 + k2ξ2
But kξ = (τω0)(ω/ω0) = ω/Ω 1 except when—as previously remarked—ω is so large as to render the classical theory meaningless. So the factor(1 + k2ξ2)–1 can/should be abandoned. The upshot: Rayleigh scattering revertsto Thomson scattering at frequencies ω the molecular resonance frequencyω0. Physically, the charge is stimulated so briskly that it does not feel itsattachment to the slow spring, and responds like a free particle. It was torepresent this fact that the red asymptote was introduced into Figure 147.
9. Radiative decay. Suppose now that the incident light beam is abruptlyswitched off. We expect the oscillating electrona to radiate its energy away,coming finally to rest. This is the process which, as explained below, givesrise to the classical theory of spectral line shape. The radiative relaxation of aharmonically bound classical electron is governed by
xxx− τ xxx... + ω2
0xxx = 000 (500)
which is just the homogeneous counterpart of (494). Borrowing τ = k/ω0 frompage 408 and multiplying by ω0 we obtain
ω0xxx− kxxx... + ω3
0xxx = 000
which proves more convenient for the purposes at hand. Looking for solutionsof the form eiωt we find that ω must be a root of the cubic polynomial
ikω3 − ω0ω + ω30 = 0
Mathematica provides complicated closed-form descriptions of those roots,which when expanded in powers of the dimensionless parameter k become
ω1 = +ω0 + i 12ω0k − 58ω0k
2 − iω0k3 + · · ·
ω2 = −ω0 + i 12ω0k + 58ω0k
2 − iω0k3 − · · ·
ω3 = −iω0
k–1 + k − 2k3 + 7k5 − · · ·
The root ω3 we abandon as an unphysical artifact because
eiω3t = exp[ω0
k–1 + k − · · ·
t]
very rapidly blows up
414 Radiative processes
That leaves us with two linearly independent solutions
e−ω0(12 k−k3+···) t · e± iω0(1− 5
8 ω0k2+···) t
and with the implication that
xxx(t) = XXX e−12 ω0k t cos
[(ω0 − 5
8ω0k2)t
]is in excellent approximation305 a particular solution of (500), and that so alsois the function got by cos → sin. In a standard notation
= XXX e−12 Γ t cos
[(ω0 − ∆ω)t
](501)
where
Γ ≡ ω0k describes the damping coefficient
∆ω ≡ 58ω0k
2 describes a small downward frequency shift
A function of the familiar design (501) is plotted in Figure 150.
Notice that it is self-interaction, as described by the small dimensionlessparameter k, that is responsible both for the slow attenuation e−
12 Γ t and
for the slight frequency shift ∆ω, and that attenuation causes the electronicoscillation (whence also the resulting radiation) to be not quite monochromatic.Turning to the Fourier transform tables (which in this instance serve betterthan Mathematica) we find306
e−βy cosαy = (β/π)∫ ∞
0
1
(x− α)2 + β2+ 1
(x + α)2 + β2
cos yx dx
The implication is that (501) can be expressed
xxx(t) = XXX
∫ ∞
0
S(ω) cosωt dω (502.1)
S(ω) ≡ Γ2π
1
[ω − (ω0 − ∆ω)]2 + ( 12Γ )2
+ 1[ω + (ω0 − ∆ω)]2 + ( 1
2Γ )2
The second term is small even for ω = 0 and dies rapidly as ω increases. Wetherefore abandon that term, and work in the good approximation that
S(ω) ≈ Γ2π
1[ω − (ω0 − ∆ω)]2 + ( 1
2Γ )2(502.2)
305 How excellent? Mathematica suppliesω0
d2
dt2 − k d3
dt3 + ω30
e−
12 ω0k te± i(ω0− 5
8 ω0k2) t
= 0 + 0k + 0k2 − i2ω30k
3 + 10364 ω
30k
4 + i 10564 ω30k
5 − · · ·306 A. Erdelyi et al (editors), Tables of Integral Transforms (), Volume I,Table 1.2#13 (page 8) and Table 1.6#19 (page 21).
Classical theory of spectral line shape 415
1
5
10
1
2π
Figure 150: Above: diagram of the motion of a charge-on-a-spring(Rayleigh’s “classical molecule”) that, because it experiences periodicacceleration, slowly radiates away its initial store of energy. Thefigure derives from (501) with ω0 = 1 and k = 0.05. The modulatingexponential factor e−
12 Γ t is shown in blue. The Fourier transform
of that curve (below) can be interpreted as a description what wouldbe seen by a physicist who examines the emitted radiation with theaid of a spectroscope. The “spectral line” has a “Lorentzian” profile.
At (502.2) we encounter once again—but this time in the frequency domain—precisely the Lorentz distribution
S(ω) ≈ L(ω − [ω0 − ∆ω], Γ )
first encountered at (498), and the basis for the statement that
Classical line shape is Lorentzian (503)
We digress to acquire familiarity with some of the basic properties of theLorentz distribution function L(x, Γ ) ≡ Γ
2π [x2 +( 12Γ )2]–1. Figure 151 shows the
416 Radiative processes
Figure 151: Characteristic shaped of what physicists usually callthe “Lorentz distribution” but mathematicians know as the “Cauchydistribution.” Arrows mark the half-max points, and Γ is shown inthe text to be the distance between those points.
characteristic shape of the Lorentz distribution. It is elementary that
L(x, Γ ) Lmax = L(0, Γ ) = 2πΓ
and thatL(x, Γ ) = 1
2Lmax =⇒ x = ± 12Γ
so the parameter Γ can be interpreted
Γ = width at half-max (504)
On casual inspection (Figure 152) the graphs of the Lorentz and Gaussian(or “normal”) distributions appear quite similar, though the former has anoticeably sharper central peak and relatively wide hips. Richard Crandall’s“The Lorentz distribution is a pig—too fat!” might seem uncharitable . . .untilone looks to the moments of the two distributions. For the Gaussian thesequence
〈x0〉, 〈x1〉, 〈x2〉, 〈x3〉, 〈x4〉, 〈x5〉, 〈x6〉, 〈x7〉, 〈x8〉, . . .proceeds unremarkably
1, 0, 12a
2, 0, 34a
4, 0, 158 a
6, 0, 10516 a
2, . . .
but in the case of the Lorentz distribution even the definition of the momentsis a bit problematic (as Mathematica is quick to remind us): if we proceed fromthe definition 〈xn〉 ≡ limz↑∞
∫ +z
−zxnL(x, Γ ) dx we obtain
1, 0, ∞, 0, ∞, 0, ∞, 0, ∞, . . .
So wide are the hips of the Lorentz distribution that (in particular)
∆x ≡√〈(x− 〈x〉)2〉 = ∞
Classical theory of spectral line shape 417
Figure 152: The Lorentz distribution L(x, Γ ) ≡ Γ2π [x2 + ( 1
2Γ )2]–1
has here been superimposed upon the Gaussian distribution
G(x, a) ≡ 1a√
πe−(x/a)2
of the same height (set a =√
π2 Γ ). The Lorentz distribution is seen
to have a relatively sharp peak, but relatively broader flanks.
The standard descriptor of the “width” of the distribution is therefore notavailable: to provide such information one is forced to adopt (504). It isremarkable that, of two distributions that—when plotted—so nearly resembleone another,
• one is arguably “the best behaved in the world,” and• the other one of the worst behaved.307
And it is in that light remarkable that in some other respects the Lorentzdistribution is quite unexceptional: for example, it leads straightforwardly to arepresentation of the δ-function
δ(x− x0) = limΓ↓0
L(x− x0, Γ ) = limε↓0
ε/π
(x− x0)2 + ε2
that often proves useful in applications. Returning now to the physics . . .
The classical theory of spectral line shape marks an interesting point inthe history of physics, but leads to results which are of enduring interest onlyas zeroth approximations to their quantum counterparts. As such, they are
307 It was known to Poisson already in that what came to be calledthe “Cauchy distribution” is a distribution to which the fundamental “centrallimit theorem” does not pertain. Cauchy himself entered the picture only in—the year of Lorentz’ birth. My source here has been the footnote thatappears on page 183 of S. M. Stigler’s The History of Statistics ().
418 Radiative processes
remarkably good. To illustrate the point: Reading from (501) we see that our“classical molecule” has a
characteristic lifetime = 2/Γ
while itsspectral linewidth = Γ/2
Evidently(linewidth)·(lifetime) = 1 (505)
Quantum mechanically, spectral line shape arises in first approximation (viaE = hν = ω) from an instance of the Heisenberg uncertainty principle,according to which
∆E · (lifetime)
But ∆E = · (linewidth) so we are, in effect, led back again to the classicalrelation (505). Similar parallels could be drawn from the quantum theory ofelectromagnetic scattering processes.308
10. Concluding remarks. Classical radiation theory, though latent in Maxwell’sequations, is a subject of which Maxwell himself knew nothing. Its developmentwas stimulated by Hertz’ experimental production/detection of electromagneticwaves—a development which Maxwell anticipated, but did not live long enoughto see—and especially by the technological effort which attended the inventionof radio. It is a subject of which we have only scratched the surface: we haveconcentrated on the radiation produced by individual accelerated charges, andremain as innocent as babies concerning the fields produced by the currentsthat flow in the antenna arrays that several generations of radio engineers haveworked so ingeniously to devise.
The subject leads, as we have witnessed, to mathematical relationshipsnotable for their complexity. But those intricate relationships among EEE ’s, BBB’s,the elements of Sµν . . . sprang from relatively simple properties of the potentialsAµ. Indeed, the work of this entire chapter (chapter in the text, chapter inthe history of pure/applied physics) can be viewed as an exercise in appliedpotential theory. It is curious that—in electrodynamics most conspicuously, butalso elsewhere in physics—it appears to be the spooks who speak the languageof God, and is in any event certainly the spooks who coordinate our effort toaccount for and describe the complexity evident in the observable/tangible worldof direct experience.
. .
Our progress thus far has (in 418 pages and ∼ 60 hours) taken us in a fairlydirect path from the “beginning” or our subject to within sight of its “end”. . . from a discussion of first principles and historical roots into the realm where
308 See, for example, W. Heitler, Quantum Theory of Radiation ().
Concluding remarks 419
electrodynamics shows an ever-stronger tendency to break down. Along theway, electrodynamics gave birth to special relativity (who has long since lefthome to lead an independent existence elsewhere) . . . and as we take leave of thelady she is clearly once again pregnant (with quantum mechanics, elementaryparticle physics, general relativity, . . . ). Her best years—if no longer as adancer, then as a teacher of dance—lie still ahead. But that is another storyfor another day. In the pages that follow we will be backtracking—discussingmiscellaneeous issues that, for all their theoretical/technological importance,were judged to be peripheral to our initial effort.